Skip to contentSkip to conversation

various memory related fixups

PATCH Superseded [email protected] 2 participants 20 messages Apr 2 – Apr 4 Library code

v1 of 10 Superseded Latest is v10 →

lore
Share & tools
lore
#0 lore raw

cover various memory related fixups

From: Randolph Sapp <[email protected]>

Nitpicks and fixes from the discovery thread on adding PocketBeagle2 support [1]. This does a lot of general setup required for the device, but these modifications themselves aren't device specific. For those specifically interested in PocketBeagle2 support and don't care about these details, my development branch is public [2].

That first patch may provoke some opinions, but honestly if that warning was still present I wouldn't have spent a week poking holes in both the EFI and LMB allocations systems. Please let me know if there is a specific usecase that it breaks though.

[1] https://lore.kernel.org/all/[email protected]/
[2] https://github.com/StaticRocket/u-boot/tree/feature/pocketbeagle2

Randolph Sapp (6):
lmb: add LMB_FDT for fdt reserved regions
efi_dt_fixup: use fdtdec_get_bool
efi_selftest_memory: check for duplicates first
efi_memory: nitpick removal loop
efi_memory: backfill EFI_CONVENTIONAL_MEMORY
memory: reserve from start_addr_sp to relocaddr

 boot/image-fdt.c                       |  5 +++-
 include/lmb.h                          |  7 +++++
 lib/efi_loader/efi_dt_fixup.c          |  3 +-
 lib/efi_loader/efi_memory.c            | 41 ++++++++++++++++++++++----
 lib/efi_selftest/efi_selftest_memory.c | 21 +++++++------
 lib/lmb.c                              | 25 +++++++++++++++-
 6 files changed, 82 insertions(+), 20 deletions(-)
Signature · 1 line
2.53.0
#1 lore raw

PATCH 1/6 lmb: add LMB_FDT for fdt reserved regions

From: Randolph Sapp <[email protected]>

Add an LMB_FDT bit for fdt reserved regions, so we can reclaim them when parsing a new device tree and properly warn people when a reservation overlaps with an existing allocation.

If we don't at least warn the user of these reservation failures, there's a chance that this region could be freed and reallocated for something important later.

This useful warning mechanism was broken in:
5a6aa7d5913 ("boot: fdt: Handle already reserved memory in boot_fdt_reserve_region()")

---
 boot/image-fdt.c |  5 ++++-
 include/lmb.h    |  7 +++++++
 lib/lmb.c        | 23 +++++++++++++++++++++++
 3 files changed, 34 insertions(+), 1 deletion(-)
boot/image-fdt.c +4 -1
index a3a4fb8b558..0f5857f24d2 100644
@@ -73,6 +73,7 @@ static void boot_fdt_reserve_region(u64 addr, u64 size, u32 flags)
 {
 	long ret;
 	phys_addr_t rsv_addr;
+	flags |= LMB_FDT;
 
 	rsv_addr = (phys_addr_t)addr;
 	ret = lmb_alloc_mem(LMB_MEM_ALLOC_ADDR, 0, &rsv_addr, size, flags);
@@ -80,7 +81,7 @@ static void boot_fdt_reserve_region(u64 addr, u64 size, u32 flags)
 		debug("   reserving fdt memory region: addr=%llx size=%llx flags=%x\n",
 		      (unsigned long long)addr,
 		      (unsigned long long)size, flags);
-	} else if (ret != -EEXIST && ret != -EINVAL) {
+	} else if (ret != -EINVAL) {
 		puts("ERROR: reserving fdt memory region failed ");
 		printf("(addr=%llx size=%llx flags=%x)\n",
 		       (unsigned long long)addr,
@@ -108,6 +109,8 @@ void boot_fdt_add_mem_rsv_regions(void *fdt_blob)
 	if (fdt_check_header(fdt_blob) != 0)
 		return;
 
+	lmb_free_fdt_regions();
+
 	/* process memreserve sections */
 	total = fdt_num_mem_rsv(fdt_blob);
 	for (i = 0; i < total; i++) {
include/lmb.h +7 -0
index 5d5f037ccb9..1b59a63f61d 100644
@@ -25,11 +25,13 @@
  * %LMB_NOMAP: Don't add to MMU configuration
  * %LMB_NOOVERWRITE: The memory region cannot be overwritten/re-reserved
  * %LMB_NONOTIFY: Do not notify other modules of changes to this memory region
+ * %LMB_FDT: A FDT reserved region that can be reclaimed if the FDT changes
  */
 #define LMB_NONE 0
 #define LMB_NOMAP BIT(1)
 #define LMB_NOOVERWRITE BIT(2)
 #define LMB_NONOTIFY BIT(3)
+#define LMB_FDT BIT(4)
 
 /**
  * enum lmb_mem_type - type of memory allocation request
@@ -215,6 +217,11 @@ phys_addr_t io_lmb_alloc(struct lmb *io_lmb, phys_size_t size, ulong align);
  */
 long io_lmb_free(struct lmb *io_lmb, phys_addr_t base, phys_size_t size);
 
+/**
+ * lmb_free_fdt_regions() - Reclaim all LMB_FDT tagged reserved regions
+ */
+void lmb_free_fdt_regions(void);
+
 #endif /* __KERNEL__ */
 
 #endif /* _LINUX_LMB_H */
lib/lmb.c +23 -0
index e2d9fe86c14..542bb11dcf5 100644
@@ -669,6 +669,29 @@ long lmb_free(phys_addr_t base, phys_size_t size, u32 flags)
 	return lmb_map_update_notify(base, size, LMB_MAP_OP_FREE, flags);
 }
 
+void lmb_free_fdt_regions(void)
+{
+	struct alist *lmb_rgn_lst = &lmb.used_mem;
+	struct lmb_region *rgn = lmb_rgn_lst->data;
+	long ret;
+
+	for (int i = 0; i < lmb_rgn_lst->count; i++) {
+		phys_addr_t base = rgn[i].base;
+		phys_size_t size = rgn[i].size;
+		u32 flags = rgn[i].flags;
+
+		if (flags & LMB_FDT) {
+			ret = lmb_free(base, size, flags);
+			if (ret < 0) {
+				printf("Unable to free FDT memory at 0x%08lx",
+				       (ulong)base);
+				continue;
+			}
+			i--;
+		}
+	}
+}
+
 static int _lmb_alloc_base(phys_size_t size, ulong align,
 			   phys_addr_t *addr, u32 flags)
 {
Signature · 1 line
2.53.0
#2 lore raw

Re: [PATCH 1/6] lmb: add LMB_FDT for fdt reserved regions

Quoted 104 lines · rs

From: Randolph Sapp <[email protected]>

Add an LMB_FDT bit for fdt reserved regions, so we can reclaim them when parsing a new device tree and properly warn people when a reservation overlaps with an existing allocation.

If we don't at least warn the user of these reservation failures, there's a chance that this region could be freed and reallocated for something important later.

This useful warning mechanism was broken in:
5a6aa7d5913 ("boot: fdt: Handle already reserved memory in boot_fdt_reserve_region()")

---
 boot/image-fdt.c |  5 ++++-
 include/lmb.h    |  7 +++++++
 lib/lmb.c        | 23 +++++++++++++++++++++++
 3 files changed, 34 insertions(+), 1 deletion(-)
boot/image-fdt.c +4 -1
index a3a4fb8b558..0f5857f24d2 100644
@@ -73,6 +73,7 @@ static void boot_fdt_reserve_region(u64 addr, u64 size, u32 flags)
 {
 	long ret;
 	phys_addr_t rsv_addr;
+	flags |= LMB_FDT;
 
 	rsv_addr = (phys_addr_t)addr;
 	ret = lmb_alloc_mem(LMB_MEM_ALLOC_ADDR, 0, &rsv_addr, size, flags);
@@ -80,7 +81,7 @@ static void boot_fdt_reserve_region(u64 addr, u64 size, u32 flags)
 		debug("   reserving fdt memory region: addr=%llx size=%llx flags=%x\n",
 		      (unsigned long long)addr,
 		      (unsigned long long)size, flags);
-	} else if (ret != -EEXIST && ret != -EINVAL) {
+	} else if (ret != -EINVAL) {
 		puts("ERROR: reserving fdt memory region failed ");
 		printf("(addr=%llx size=%llx flags=%x)\n",
 		       (unsigned long long)addr,
@@ -108,6 +109,8 @@ void boot_fdt_add_mem_rsv_regions(void *fdt_blob)
 	if (fdt_check_header(fdt_blob) != 0)
 		return;
 
+	lmb_free_fdt_regions();
+
 	/* process memreserve sections */
 	total = fdt_num_mem_rsv(fdt_blob);
 	for (i = 0; i < total; i++) {
include/lmb.h +7 -0
index 5d5f037ccb9..1b59a63f61d 100644
@@ -25,11 +25,13 @@
  * %LMB_NOMAP: Don't add to MMU configuration
  * %LMB_NOOVERWRITE: The memory region cannot be overwritten/re-reserved
  * %LMB_NONOTIFY: Do not notify other modules of changes to this memory region
+ * %LMB_FDT: A FDT reserved region that can be reclaimed if the FDT changes
  */
 #define LMB_NONE 0
 #define LMB_NOMAP BIT(1)
 #define LMB_NOOVERWRITE BIT(2)
 #define LMB_NONOTIFY BIT(3)
+#define LMB_FDT BIT(4)
 
 /**
  * enum lmb_mem_type - type of memory allocation request
@@ -215,6 +217,11 @@ phys_addr_t io_lmb_alloc(struct lmb *io_lmb, phys_size_t size, ulong align);
  */
 long io_lmb_free(struct lmb *io_lmb, phys_addr_t base, phys_size_t size);
 
+/**
+ * lmb_free_fdt_regions() - Reclaim all LMB_FDT tagged reserved regions
+ */
+void lmb_free_fdt_regions(void);
+
 #endif /* __KERNEL__ */
 
 #endif /* _LINUX_LMB_H */
lib/lmb.c +16 -0
index e2d9fe86c14..542bb11dcf5 100644
@@ -669,6 +669,29 @@ long lmb_free(phys_addr_t base, phys_size_t size, u32 flags)
 	return lmb_map_update_notify(base, size, LMB_MAP_OP_FREE, flags);
 }
 
+void lmb_free_fdt_regions(void)
+{
+	struct alist *lmb_rgn_lst = &lmb.used_mem;
+	struct lmb_region *rgn = lmb_rgn_lst->data;
+	long ret;
+
+	for (int i = 0; i < lmb_rgn_lst->count; i++) {
+		phys_addr_t base = rgn[i].base;
+		phys_size_t size = rgn[i].size;
+		u32 flags = rgn[i].flags;
+
+		if (flags & LMB_FDT) {
+			ret = lmb_free(base, size, flags);
+			if (ret < 0) {
+				printf("Unable to free FDT memory at 0x%08lx",
+				       (ulong)base);

lib/lmb.c long lmb_free(phys_addr_t base, phys_size_t size, u32 flags) · line 697

Quoted 10 lines
lib/lmb.c +7 -0
+				continue;
+			}
+			i--;
+		}
+	}
+}
+
 static int _lmb_alloc_base(phys_size_t size, ulong align,
 			   phys_addr_t *addr, u32 flags)
 {

Ugh. Looking at that loop in lmb_free_fdt_regions more I really should have written it differently to avoid the additional increments and decrements. I'll fix that in v2.

#3 lore raw

PATCH 2/6 efi_dt_fixup: use fdtdec_get_bool

In reply to [email protected]

From: Randolph Sapp <[email protected]>

Use the more straightforward fdtdec_get_bool instead of fdt_getprop and a return code check.

---
 lib/efi_loader/efi_dt_fixup.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
lib/efi_loader/efi_dt_fixup.c +1 -2
index 544e1aa9808..333711b9957 100644
@@ -123,8 +123,7 @@ void efi_carve_out_dt_rsv(void *fdt)
 			    fdtdec_get_is_enabled(fdt, subnode)) {
 				bool nomap;
 
-				nomap = !!fdt_getprop(fdt, subnode, "no-map",
-						      NULL);
+				nomap = fdtdec_get_bool(fdt, subnode, "no-map");
 				efi_reserve_memory(fdt_addr, fdt_size, nomap);
 			}
 			subnode = fdt_next_subnode(fdt, subnode);
Signature · 1 line
2.53.0
Reviewed-by on v1 2/6
#4 lore raw

Re: [PATCH 2/6] efi_dt_fixup: use fdtdec_get_bool

Quoted 8 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

Use the more straightforward fdtdec_get_bool instead of fdt_getprop and a return code check.

---

Quoted 20 lines
 lib/efi_loader/efi_dt_fixup.c | 3 +--
 1 file changed, 1 insertion(+), 2 deletions(-)
lib/efi_loader/efi_dt_fixup.c +1 -3
index 544e1aa9808..333711b9957 100644
@@ -123,8 +123,7 @@ void efi_carve_out_dt_rsv(void *fdt)
                            fdtdec_get_is_enabled(fdt, subnode)) {
                                bool nomap;
 
-                               nomap = !!fdt_getprop(fdt, subnode, "no-map",
-                                                     NULL);
+                               nomap = fdtdec_get_bool(fdt, subnode, "no-map");
                                efi_reserve_memory(fdt_addr, fdt_size, nomap);
                        }
                        subnode = fdt_next_subnode(fdt, subnode);
--

2.53.0

#5 lore raw

PATCH 3/6 efi_selftest_memory: check for duplicates first

In reply to [email protected]

From: Randolph Sapp <[email protected]>

Check for duplicate memory mappings before reporting any incorrect attributes. Could be that second allocation has the correct type while the first doesn't. Knowing there is a duplicate in this scenario is more helpful than just reporting the first mismatch.

---
 lib/efi_selftest/efi_selftest_memory.c | 21 ++++++++++-----------
 1 file changed, 10 insertions(+), 11 deletions(-)
lib/efi_selftest/efi_selftest_memory.c +10 -11
index 4d32a280061..7320964c129 100644
@@ -60,7 +60,7 @@ static int find_in_memory_map(efi_uintn_t map_size,
 			      u64 addr, int memory_type)
 {
 	efi_uintn_t i;
-	bool found = false;
+	struct efi_mem_desc *match = NULL;
 
 	for (i = 0; map_size; ++i, map_size -= desc_size) {
 		struct efi_mem_desc *entry = &memory_map[i];
@@ -72,24 +72,23 @@ static int find_in_memory_map(efi_uintn_t map_size,
 
 		if (addr >= entry->physical_start &&
 		    addr < entry->physical_start +
-			    (entry->num_pages << EFI_PAGE_SHIFT)) {
-			if (found) {
+				    (entry->num_pages << EFI_PAGE_SHIFT)) {
+			if (match) {
 				efi_st_error("Duplicate memory map entry\n");
 				return EFI_ST_FAILURE;
 			}
-			found = true;
-			if (memory_type != entry->type) {
-				efi_st_error
-					("Wrong memory type %d, expected %d\n",
-					 entry->type, memory_type);
-				return EFI_ST_FAILURE;
-			}
+			match = entry;
 		}
 	}
-	if (!found) {
+	if (!match) {
 		efi_st_error("Missing memory map entry\n");
 		return EFI_ST_FAILURE;
 	}
+	if (memory_type != match->type) {
+		efi_st_error("Wrong memory type %d, expected %d\n", match->type,
+			     memory_type);
+		return EFI_ST_FAILURE;
+	}
 	return EFI_ST_SUCCESS;
 }
 
Signature · 1 line
2.53.0
#6 lore raw

PATCH 4/6 efi_memory: nitpick removal loop

In reply to [email protected]

From: Randolph Sapp <[email protected]>

I don't even want to think about the possibility of this pointer containing a reference to something from a previous iteration.

---
 lib/efi_loader/efi_memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/efi_loader/efi_memory.c +1 -1
index b77c2f980cc..882366a9f8a 100644
@@ -129,13 +129,13 @@ static uint64_t desc_get_end(struct efi_mem_desc *desc)
 static void efi_mem_sort(void)
 {
 	struct efi_mem_list *lmem;
-	struct efi_mem_list *prevmem = NULL;
 	bool merge_again = true;
 
 	list_sort(NULL, &efi_mem, efi_mem_cmp);
 
 	/* Now merge entries that can be merged */
 	while (merge_again) {
+		struct efi_mem_list *prevmem = NULL;
 		merge_again = false;
 		list_for_each_entry(lmem, &efi_mem, link) {
 			struct efi_mem_desc *prev;
Signature · 1 line
2.53.0
#7 lore raw

Re: [PATCH 4/6] efi_memory: nitpick removal loop

I don't mind merging this, but why is the current code flawed?

Looking at the function even if prevmem contains a previous valid entry the check will fail

if ((desc_get_end(cur) == prev->physical_start) &&

    (prev->type == cur->type) &&
    (prev->attribute == cur->attribute)) {

and re-assign prevmem.

So please adjust the commit message if you want this merged.

Thanks
/Ilias

Quoted 33 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

I don't even want to think about the possibility of this pointer containing a reference to something from a previous iteration.

---
 lib/efi_loader/efi_memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/efi_loader/efi_memory.c +1 -2
index b77c2f980cc..882366a9f8a 100644
@@ -129,13 +129,13 @@ static uint64_t desc_get_end(struct efi_mem_desc *desc)
 static void efi_mem_sort(void)
 {
        struct efi_mem_list *lmem;
-       struct efi_mem_list *prevmem = NULL;
        bool merge_again = true;
 
        list_sort(NULL, &efi_mem, efi_mem_cmp);
 
        /* Now merge entries that can be merged */
        while (merge_again) {
+               struct efi_mem_list *prevmem = NULL;
                merge_again = false;
                list_for_each_entry(lmem, &efi_mem, link) {
                        struct efi_mem_desc *prev;
--

2.53.0

#8 lore raw

Re: [PATCH 4/6] efi_memory: nitpick removal loop

Quoted 9 lines · Ilias Apalodimas

I don't mind merging this, but why is the current code flawed?

Looking at the function even if prevmem contains a previous valid entry the check will fail

if ((desc_get_end(cur) == prev->physical_start) &&

    (prev->type == cur->type) &&
    (prev->attribute == cur->attribute)) {
Quoted 6 lines

and re-assign prevmem.

So please adjust the commit message if you want this merged.

Thanks
/Ilias

Ah, not really that it is "flawed" right now, but more so that we should never even have to think about that occurrence, because at best it's an unnecessary compare. Why waste time trying to defend or attack a line of code that simply shouldn't exist?

Quoted 34 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:
Quoted 33 lines

From: Randolph Sapp <[email protected]>

I don't even want to think about the possibility of this pointer containing a reference to something from a previous iteration.

---
 lib/efi_loader/efi_memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/efi_loader/efi_memory.c +1 -2
index b77c2f980cc..882366a9f8a 100644
@@ -129,13 +129,13 @@ static uint64_t desc_get_end(struct efi_mem_desc *desc)
 static void efi_mem_sort(void)
 {
        struct efi_mem_list *lmem;
-       struct efi_mem_list *prevmem = NULL;
        bool merge_again = true;
 
        list_sort(NULL, &efi_mem, efi_mem_cmp);
 
        /* Now merge entries that can be merged */
        while (merge_again) {
+               struct efi_mem_list *prevmem = NULL;
                merge_again = false;
                list_for_each_entry(lmem, &efi_mem, link) {
                        struct efi_mem_desc *prev;
--

2.53.0

#9 lore raw

Re: [PATCH 4/6] efi_memory: nitpick removal loop

Quoted 15 lines · Randolph Sapp
Quoted 13 lines · Ilias Apalodimas

I don't mind merging this, but why is the current code flawed?

Looking at the function even if prevmem contains a previous valid entry the check will fail

if ((desc_get_end(cur) == prev->physical_start) &&

    (prev->type == cur->type) &&
    (prev->attribute == cur->attribute)) {

and re-assign prevmem.

So please adjust the commit message if you want this merged.

Quoted 7 lines
Quoted 2 lines

Thanks
/Ilias

Ah, not really that it is "flawed" right now, but more so that we should never even have to think about that occurrence, because at best it's an unnecessary compare. Why waste time trying to defend or attack a line of code that simply shouldn't exist?

Not attacking anything. But the commit message wasn't as clear and I ended up looking all the code. Chances are this is going to happen if someone reads that commit message in a few moths. So something along the lines of "This not currently a problem but ...." etc is what we should have in that commit.

Cheers
/Ilias

Quoted 36 lines
Quoted 35 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:
Quoted 33 lines

From: Randolph Sapp <[email protected]>

I don't even want to think about the possibility of this pointer containing a reference to something from a previous iteration.

---
 lib/efi_loader/efi_memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/efi_loader/efi_memory.c +1 -2
index b77c2f980cc..882366a9f8a 100644
@@ -129,13 +129,13 @@ static uint64_t desc_get_end(struct efi_mem_desc *desc)
 static void efi_mem_sort(void)
 {
        struct efi_mem_list *lmem;
-       struct efi_mem_list *prevmem = NULL;
        bool merge_again = true;
 
        list_sort(NULL, &efi_mem, efi_mem_cmp);
 
        /* Now merge entries that can be merged */
        while (merge_again) {
+               struct efi_mem_list *prevmem = NULL;
                merge_again = false;
                list_for_each_entry(lmem, &efi_mem, link) {
                        struct efi_mem_desc *prev;
--

2.53.0

#10 lore raw

Re: [PATCH 4/6] efi_memory: nitpick removal loop

Quoted 24 lines · Ilias Apalodimas
Quoted 15 lines · Randolph Sapp
Quoted 13 lines · Ilias Apalodimas

I don't mind merging this, but why is the current code flawed?

Looking at the function even if prevmem contains a previous valid entry the check will fail

if ((desc_get_end(cur) == prev->physical_start) &&

    (prev->type == cur->type) &&
    (prev->attribute == cur->attribute)) {

and re-assign prevmem.

So please adjust the commit message if you want this merged.

Quoted 7 lines
Quoted 2 lines

Thanks
/Ilias

Ah, not really that it is "flawed" right now, but more so that we should never even have to think about that occurrence, because at best it's an unnecessary compare. Why waste time trying to defend or attack a line of code that simply shouldn't exist?

Quoted 8 lines

Not attacking anything. But the commit message wasn't as clear and I ended up looking all the code. Chances are this is going to happen if someone reads that commit message in a few moths. So something along the lines of "This not currently a problem but ...." etc is what we should have in that commit.

Cheers
/Ilias

I didn't mean for it to sound as if you were attacking it. Change requests (like this patch) are always the "attack." I'm just being lazy and saying I didn't want to parse this and come up with some specific scenario where it could be a problem.

I'll adjust the description to indicate the *usually* unnecessary comparison that's occurring right now. Should be a good enough justification on it's own.

Quoted 36 lines
Quoted 36 lines
Quoted 35 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:
Quoted 33 lines

From: Randolph Sapp <[email protected]>

I don't even want to think about the possibility of this pointer containing a reference to something from a previous iteration.

---
 lib/efi_loader/efi_memory.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
lib/efi_loader/efi_memory.c +1 -2
index b77c2f980cc..882366a9f8a 100644
@@ -129,13 +129,13 @@ static uint64_t desc_get_end(struct efi_mem_desc *desc)
 static void efi_mem_sort(void)
 {
        struct efi_mem_list *lmem;
-       struct efi_mem_list *prevmem = NULL;
        bool merge_again = true;
 
        list_sort(NULL, &efi_mem, efi_mem_cmp);
 
        /* Now merge entries that can be merged */
        while (merge_again) {
+               struct efi_mem_list *prevmem = NULL;
                merge_again = false;
                list_for_each_entry(lmem, &efi_mem, link) {
                        struct efi_mem_desc *prev;
--

2.53.0

#11 lore raw

PATCH 5/6 efi_memory: backfill EFI_CONVENTIONAL_MEMORY

In reply to [email protected]

From: Randolph Sapp <[email protected]>

Backfill any fragmented EFI_CONVENTIONAL_MEMORY when LMB allocations start to fail. This may be a low memory environment, or maybe LMB is running up against a tricky reserved region.

---
 lib/efi_loader/efi_memory.c | 35 +++++++++++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)
lib/efi_loader/efi_memory.c +33 -2
index 882366a9f8a..f07cc39b157 100644
@@ -441,6 +441,27 @@ static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated)
 	return EFI_NOT_FOUND;
 }
 
+/**
+ * efi_get_conventional_start - find the highest conventional region start
+ *                              address with at least the specified number of
+ *                              pages
+ *
+ * @pages:                      number of pages required to be in that carveout
+ * Return:                      starting address of the give carveout
+ */
+static u64 efi_get_conventional_start(u64 pages)
+{
+	struct efi_mem_list *item;
+
+	list_for_each_entry(item, &efi_mem, link) {
+		if (item->desc.type != EFI_CONVENTIONAL_MEMORY)
+			continue;
+		if (item->desc.num_pages >= pages)
+			return item->desc.physical_start;
+	}
+	return 0;
+}
+
 /**
  * efi_allocate_pages - allocate memory pages
  *
@@ -507,10 +528,20 @@ efi_status_t efi_allocate_pages(enum efi_allocate_type type,
 	/* Reserve that map in our memory maps */
 	ret = efi_update_memory_map(efi_addr, pages, memory_type, true, false);
 	if (ret != EFI_SUCCESS) {
-		/* Map would overlap, bail out */
+		/* Map would overlap, try something else */
 		lmb_free(addr, (u64)pages << EFI_PAGE_SHIFT, flags);
 		unmap_sysmem((void *)(uintptr_t)efi_addr);
-		return  EFI_OUT_OF_RESOURCES;
+
+		/* See if there is any EFI_CONVENTIONAL_MEMORY allocations */
+		if (type != EFI_ALLOCATE_ADDRESS) {
+			*memory = efi_get_conventional_start(pages);
+			if (*memory != 0)
+				return efi_allocate_pages(EFI_ALLOCATE_ADDRESS,
+							  memory_type, pages,
+							  memory);
+		}
+
+		return EFI_OUT_OF_RESOURCES;
 	}
 
 	*memory = efi_addr;
Signature · 1 line
2.53.0
#12 lore raw

Re: [PATCH 5/6] efi_memory: backfill EFI_CONVENTIONAL_MEMORY

Hi Randolph

Quoted 29 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

Backfill any fragmented EFI_CONVENTIONAL_MEMORY when LMB allocations start to fail. This may be a low memory environment, or maybe LMB is running up against a tricky reserved region.

---
 lib/efi_loader/efi_memory.c | 35 +++++++++++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)
lib/efi_loader/efi_memory.c +9 -0
index 882366a9f8a..f07cc39b157 100644
@@ -441,6 +441,27 @@ static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated)
        return EFI_NOT_FOUND;
 }
 
+/**
+ * efi_get_conventional_start - find the highest conventional region start
+ *                              address with at least the specified number of
+ *                              pages
+ *
+ * @pages:                      number of pages required to be in that carveout
+ * Return:                      starting address of the give carveout
+ */
+static u64 efi_get_conventional_start(u64 pages)

lib/efi_loader/efi_memory.c static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated) · line 462

Quoted 10 lines
lib/efi_loader/efi_memory.c +10 -0
+{
+       struct efi_mem_list *item;
+
+       list_for_each_entry(item, &efi_mem, link) {
+               if (item->desc.type != EFI_CONVENTIONAL_MEMORY)
+                       continue;
+               if (item->desc.num_pages >= pages)
+                       return item->desc.physical_start;
+       }
+       return 0;

We've been bitten by this in the past and although unlikely 0 is a valid address for some hardware. Can we return something different if we don't find a match? UINT64_MAX perhaps?

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 529

Quoted 8 lines
lib/efi_loader/efi_memory.c +2 -0
+}
+
 /**
  * efi_allocate_pages - allocate memory pages
  *
@@ -507,10 +528,20 @@ efi_status_t efi_allocate_pages(enum efi_allocate_type type,
        /* Reserve that map in our memory maps */
        ret = efi_update_memory_map(efi_addr, pages, memory_type, true, false);

In theory the lmb and the EFI maps are in sync. I haven't checked close enough yet, but do you have cases where lmb_alloc worked and updating the efi memory map failed?

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 536

Quoted 9 lines
lib/efi_loader/efi_memory.c +4 -2
        if (ret != EFI_SUCCESS) {
-               /* Map would overlap, bail out */
+               /* Map would overlap, try something else */
                lmb_free(addr, (u64)pages << EFI_PAGE_SHIFT, flags);
                unmap_sysmem((void *)(uintptr_t)efi_addr);
-               return  EFI_OUT_OF_RESOURCES;
+
+               /* See if there is any EFI_CONVENTIONAL_MEMORY allocations */
+               if (type != EFI_ALLOCATE_ADDRESS) {

Can you please inverse this. It's going to reduce the identation. if (type == EFI_ALLOCATE_ADDRESS)

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 538

Quoted 2 lines
lib/efi_loader/efi_memory.c +2 -0
+                       *memory = efi_get_conventional_start(pages);
+                       if (*memory != 0)

Same here

Quoted 10 lines
lib/efi_loader/efi_memory.c +6 -1
+                               return efi_allocate_pages(EFI_ALLOCATE_ADDRESS,
+                                                         memory_type, pages,
+                                                         memory);
+               }
+
+               return EFI_OUT_OF_RESOURCES;
        }
 
        *memory = efi_addr;
--
Quoted 2 lines

2.53.0

Thanks
/Ilias

#13 lore raw

Re: [PATCH 5/6] efi_memory: backfill EFI_CONVENTIONAL_MEMORY

Quoted 47 lines · Ilias Apalodimas

Hi Randolph

Quoted 29 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

Backfill any fragmented EFI_CONVENTIONAL_MEMORY when LMB allocations start to fail. This may be a low memory environment, or maybe LMB is running up against a tricky reserved region.

---
 lib/efi_loader/efi_memory.c | 35 +++++++++++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)
lib/efi_loader/efi_memory.c +9 -0
index 882366a9f8a..f07cc39b157 100644
@@ -441,6 +441,27 @@ static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated)
        return EFI_NOT_FOUND;
 }
 
+/**
+ * efi_get_conventional_start - find the highest conventional region start
+ *                              address with at least the specified number of
+ *                              pages
+ *
+ * @pages:                      number of pages required to be in that carveout
+ * Return:                      starting address of the give carveout
+ */
+static u64 efi_get_conventional_start(u64 pages)

lib/efi_loader/efi_memory.c static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated) · line 462

Quoted 10 lines
lib/efi_loader/efi_memory.c +10 -0
+{
+       struct efi_mem_list *item;
+
+       list_for_each_entry(item, &efi_mem, link) {
+               if (item->desc.type != EFI_CONVENTIONAL_MEMORY)
+                       continue;
+               if (item->desc.num_pages >= pages)
+                       return item->desc.physical_start;
+       }
+       return 0;

We've been bitten by this in the past and although unlikely 0 is a valid address for some hardware. Can we return something different if we don't find a match? UINT64_MAX perhaps?

Quoted 12 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 529

Quoted 8 lines
lib/efi_loader/efi_memory.c +2 -0
+}
+
 /**
  * efi_allocate_pages - allocate memory pages
  *
@@ -507,10 +528,20 @@ efi_status_t efi_allocate_pages(enum efi_allocate_type type,
        /* Reserve that map in our memory maps */
        ret = efi_update_memory_map(efi_addr, pages, memory_type, true, false);

In theory the lmb and the EFI maps are in sync. I haven't checked close enough yet, but do you have cases where lmb_alloc worked and updating the efi memory map failed?

Fair point. Now that I've fixed the FDT warning I should probably check if I can actually reproduce the issue that actually required this.

Quoted 32 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 536

Quoted 9 lines
lib/efi_loader/efi_memory.c +4 -2
        if (ret != EFI_SUCCESS) {
-               /* Map would overlap, bail out */
+               /* Map would overlap, try something else */
                lmb_free(addr, (u64)pages << EFI_PAGE_SHIFT, flags);
                unmap_sysmem((void *)(uintptr_t)efi_addr);
-               return  EFI_OUT_OF_RESOURCES;
+
+               /* See if there is any EFI_CONVENTIONAL_MEMORY allocations */
+               if (type != EFI_ALLOCATE_ADDRESS) {

Can you please inverse this. It's going to reduce the identation. if (type == EFI_ALLOCATE_ADDRESS)

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 538

Quoted 2 lines
lib/efi_loader/efi_memory.c +2 -0
+                       *memory = efi_get_conventional_start(pages);
+                       if (*memory != 0)

Same here

Quoted 10 lines
lib/efi_loader/efi_memory.c +6 -1
+                               return efi_allocate_pages(EFI_ALLOCATE_ADDRESS,
+                                                         memory_type, pages,
+                                                         memory);
+               }
+
+               return EFI_OUT_OF_RESOURCES;
        }
 
        *memory = efi_addr;
--
Quoted 2 lines

2.53.0

Thanks
/Ilias

#14 lore raw

Re: [PATCH 5/6] efi_memory: backfill EFI_CONVENTIONAL_MEMORY

Quoted 61 lines · Randolph Sapp
Quoted 47 lines · Ilias Apalodimas

Hi Randolph

Quoted 29 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

Backfill any fragmented EFI_CONVENTIONAL_MEMORY when LMB allocations start to fail. This may be a low memory environment, or maybe LMB is running up against a tricky reserved region.

---
 lib/efi_loader/efi_memory.c | 35 +++++++++++++++++++++++++++++++++--
 1 file changed, 33 insertions(+), 2 deletions(-)
lib/efi_loader/efi_memory.c +9 -0
index 882366a9f8a..f07cc39b157 100644
@@ -441,6 +441,27 @@ static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated)
        return EFI_NOT_FOUND;
 }
 
+/**
+ * efi_get_conventional_start - find the highest conventional region start
+ *                              address with at least the specified number of
+ *                              pages
+ *
+ * @pages:                      number of pages required to be in that carveout
+ * Return:                      starting address of the give carveout
+ */
+static u64 efi_get_conventional_start(u64 pages)

lib/efi_loader/efi_memory.c static efi_status_t efi_check_allocated(u64 addr, bool must_be_allocated) · line 462

Quoted 10 lines
lib/efi_loader/efi_memory.c +10 -0
+{
+       struct efi_mem_list *item;
+
+       list_for_each_entry(item, &efi_mem, link) {
+               if (item->desc.type != EFI_CONVENTIONAL_MEMORY)
+                       continue;
+               if (item->desc.num_pages >= pages)
+                       return item->desc.physical_start;
+       }
+       return 0;

We've been bitten by this in the past and although unlikely 0 is a valid address for some hardware. Can we return something different if we don't find a match? UINT64_MAX perhaps?

Quoted 12 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 529

Quoted 8 lines
lib/efi_loader/efi_memory.c +2 -0
+}
+
 /**
  * efi_allocate_pages - allocate memory pages
  *
@@ -507,10 +528,20 @@ efi_status_t efi_allocate_pages(enum efi_allocate_type type,
        /* Reserve that map in our memory maps */
        ret = efi_update_memory_map(efi_addr, pages, memory_type, true, false);

In theory the lmb and the EFI maps are in sync. I haven't checked close enough yet, but do you have cases where lmb_alloc worked and updating the efi memory map failed?

Quoted 2 lines

Fair point. Now that I've fixed the FDT warning I should probably check if I can actually reproduce the issue that actually required this.

Yeah, this patch isn't necessary anymore. I suppose it should be dropped as it would potentially mask any discrepancies between the two maps.

Out of curiosity, is there any plan to merge the EFI and LMB allocators more in the future, or is the current layering scheme the furthest we want to go?

Quoted 32 lines
Quoted 32 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 536

Quoted 9 lines
lib/efi_loader/efi_memory.c +4 -2
        if (ret != EFI_SUCCESS) {
-               /* Map would overlap, bail out */
+               /* Map would overlap, try something else */
                lmb_free(addr, (u64)pages << EFI_PAGE_SHIFT, flags);
                unmap_sysmem((void *)(uintptr_t)efi_addr);
-               return  EFI_OUT_OF_RESOURCES;
+
+               /* See if there is any EFI_CONVENTIONAL_MEMORY allocations */
+               if (type != EFI_ALLOCATE_ADDRESS) {

Can you please inverse this. It's going to reduce the identation. if (type == EFI_ALLOCATE_ADDRESS)

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 538

Quoted 2 lines
lib/efi_loader/efi_memory.c +2 -0
+                       *memory = efi_get_conventional_start(pages);
+                       if (*memory != 0)

Same here

Quoted 10 lines
lib/efi_loader/efi_memory.c +6 -1
+                               return efi_allocate_pages(EFI_ALLOCATE_ADDRESS,
+                                                         memory_type, pages,
+                                                         memory);
+               }
+
+               return EFI_OUT_OF_RESOURCES;
        }
 
        *memory = efi_addr;
--
Quoted 2 lines

2.53.0

Thanks
/Ilias

#15 lore raw

Re: [PATCH 5/6] efi_memory: backfill EFI_CONVENTIONAL_MEMORY

[...]

Quoted 14 lines
Quoted 14 lines
Quoted 13 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 529

Quoted 9 lines
lib/efi_loader/efi_memory.c +2 -0
+}
+
 /**
  * efi_allocate_pages - allocate memory pages
  *
@@ -507,10 +528,20 @@ efi_status_t efi_allocate_pages(enum efi_allocate_type type,
        /* Reserve that map in our memory maps */
        ret = efi_update_memory_map(efi_addr, pages, memory_type, true, false);

In theory the lmb and the EFI maps are in sync. I haven't checked close enough yet, but do you have cases where lmb_alloc worked and updating the efi memory map failed?

Quoted 5 lines
Quoted 2 lines

Fair point. Now that I've fixed the FDT warning I should probably check if I can actually reproduce the issue that actually required this.

Yeah, this patch isn't necessary anymore. I suppose it should be dropped as it would potentially mask any discrepancies between the two maps.

Yea and if we really need a similar fix in the future, it's best if we put in the lmb core code.

Quoted 3 lines

Out of curiosity, is there any plan to merge the EFI and LMB allocators more in the future, or is the current layering scheme the furthest we want to go?

We do, but havent due to lack of time. It's been a while since I discussed this with Heinrich but iirc the only thing missing is for lmb to be aware of the memory flags EFI expects.

/Ilias

Quoted 34 lines
Quoted 33 lines
Quoted 33 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 536

Quoted 10 lines
lib/efi_loader/efi_memory.c +4 -2
        if (ret != EFI_SUCCESS) {
-               /* Map would overlap, bail out */
+               /* Map would overlap, try something else */
                lmb_free(addr, (u64)pages << EFI_PAGE_SHIFT, flags);
                unmap_sysmem((void *)(uintptr_t)efi_addr);
-               return  EFI_OUT_OF_RESOURCES;
+
+               /* See if there is any EFI_CONVENTIONAL_MEMORY allocations */
+               if (type != EFI_ALLOCATE_ADDRESS) {

Can you please inverse this. It's going to reduce the identation. if (type == EFI_ALLOCATE_ADDRESS)

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 538

Quoted 2 lines
lib/efi_loader/efi_memory.c +2 -0
+                       *memory = efi_get_conventional_start(pages);
+                       if (*memory != 0)

Same here

Quoted 10 lines
lib/efi_loader/efi_memory.c +6 -1
+                               return efi_allocate_pages(EFI_ALLOCATE_ADDRESS,
+                                                         memory_type, pages,
+                                                         memory);
+               }
+
+               return EFI_OUT_OF_RESOURCES;
        }
 
        *memory = efi_addr;
--
Quoted 2 lines

2.53.0

Thanks
/Ilias

#16 lore raw

Re: [PATCH 5/6] efi_memory: backfill EFI_CONVENTIONAL_MEMORY

Quoted 26 lines · Ilias Apalodimas

[...]

Quoted 14 lines
Quoted 14 lines
Quoted 13 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 529

Quoted 9 lines
lib/efi_loader/efi_memory.c +2 -0
+}
+
 /**
  * efi_allocate_pages - allocate memory pages
  *
@@ -507,10 +528,20 @@ efi_status_t efi_allocate_pages(enum efi_allocate_type type,
        /* Reserve that map in our memory maps */
        ret = efi_update_memory_map(efi_addr, pages, memory_type, true, false);

In theory the lmb and the EFI maps are in sync. I haven't checked close enough yet, but do you have cases where lmb_alloc worked and updating the efi memory map failed?

Quoted 5 lines
Quoted 2 lines

Fair point. Now that I've fixed the FDT warning I should probably check if I can actually reproduce the issue that actually required this.

Yeah, this patch isn't necessary anymore. I suppose it should be dropped as it would potentially mask any discrepancies between the two maps.

Yea and if we really need a similar fix in the future, it's best if we put in the lmb core code.

Quoted 1 line
Quoted 8 lines
Quoted 2 lines

Out of curiosity, is there any plan to merge the EFI and LMB allocators more in the future, or is the current layering scheme the furthest we want to go?

We do, but havent due to lack of time. It's been a while since I discussed this with Heinrich but iirc the only thing missing is for lmb to be aware of the memory flags EFI expects.

/Ilias

I mean, trimming the last x bits off of the LMB allocation flag value to store the current EFI enum is a little bit of a hack, but honestly it'll allow the current allocation coalescence logic to continue to work as-is. It would just need helpers to store and retrieve the enum type for the code that actually cares about the EFI memory types. I could look into doing that if you guys think that's a valid approach, or if you've got other ideas.

Quoted 34 lines
Quoted 34 lines
Quoted 33 lines
Quoted 33 lines

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 536

Quoted 10 lines
lib/efi_loader/efi_memory.c +4 -2
        if (ret != EFI_SUCCESS) {
-               /* Map would overlap, bail out */
+               /* Map would overlap, try something else */
                lmb_free(addr, (u64)pages << EFI_PAGE_SHIFT, flags);
                unmap_sysmem((void *)(uintptr_t)efi_addr);
-               return  EFI_OUT_OF_RESOURCES;
+
+               /* See if there is any EFI_CONVENTIONAL_MEMORY allocations */
+               if (type != EFI_ALLOCATE_ADDRESS) {

Can you please inverse this. It's going to reduce the identation. if (type == EFI_ALLOCATE_ADDRESS)

lib/efi_loader/efi_memory.c efi_status_t efi_allocate_pages(enum efi_allocate_type type, · line 538

Quoted 2 lines
lib/efi_loader/efi_memory.c +2 -0
+                       *memory = efi_get_conventional_start(pages);
+                       if (*memory != 0)

Same here

Quoted 10 lines
lib/efi_loader/efi_memory.c +6 -1
+                               return efi_allocate_pages(EFI_ALLOCATE_ADDRESS,
+                                                         memory_type, pages,
+                                                         memory);
+               }
+
+               return EFI_OUT_OF_RESOURCES;
        }
 
        *memory = efi_addr;
--
Quoted 2 lines

2.53.0

Thanks
/Ilias

#17 lore raw

PATCH 6/6 memory: reserve from start_addr_sp to relocaddr

In reply to [email protected]

From: Randolph Sapp <[email protected]>

Reserve the memory from gd->start_addr_sp - CONFIG_STACK_SIZE to gd->relocaddr instead of gd->ram_top. This allows platform specific relocation addresses to work without unnecessarily painting over a large range.

---
 lib/efi_loader/efi_memory.c | 4 ++--
 lib/lmb.c                   | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
lib/efi_loader/efi_memory.c +2 -2
index f07cc39b157..57691d15758 100644
@@ -915,8 +915,8 @@ static void add_u_boot_and_runtime(void)
 	/* Add U-Boot */
 	uboot_start = ((uintptr_t)map_sysmem(gd->start_addr_sp, 0) -
 		       uboot_stack_size) & ~EFI_PAGE_MASK;
-	uboot_pages = ((uintptr_t)map_sysmem(gd->ram_top - 1, 0) -
-		       uboot_start + EFI_PAGE_MASK) >> EFI_PAGE_SHIFT;
+	uboot_pages = ((gd->relocaddr - uboot_start) + EFI_PAGE_MASK) >>
+		      EFI_PAGE_SHIFT;
 	efi_update_memory_map(uboot_start, uboot_pages, EFI_BOOT_SERVICES_CODE,
 			      false, false);
 #if defined(__aarch64__)
lib/lmb.c +1 -1
index 542bb11dcf5..0df8157db7f 100644
@@ -540,7 +540,7 @@ static void lmb_reserve_uboot_region(void)
 	ulong pram = 0;
 
 	rsv_start = gd->start_addr_sp - CONFIG_STACK_SIZE;
-	end = gd->ram_top;
+	end = gd->relocaddr;
 
 	/*
 	 * Reserve memory from aligned address below the bottom of U-Boot stack
Signature · 1 line
2.53.0
#18 lore raw

Re: [PATCH 6/6] memory: reserve from start_addr_sp to relocaddr

Quoted 7 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

Reserve the memory from gd->start_addr_sp - CONFIG_STACK_SIZE to gd->relocaddr instead of gd->ram_top. This allows platform specific relocation addresses to work without unnecessarily painting over a large range.

This hangs when trying to boot with UEFI on QEMU aarch64. With the qemu_arm64_lwip_defconfig and this command line

qemu-system-aarch64 -m 8192 -smp 2 -nographic -cpu cortex-a57 \
    -machine virt,secure=off \
    -bios u-boot.bin \
    -device virtio-rng-pci \
    -drive id=os,if=none,file=my.iso \
    -device virtio-blk-device,drive=os \
    -object memory-backend-ram,id=ram0,size=4G \
    -object memory-backend-ram,id=ram1,size=4G \
    -numa node,memdev=ram0 \
    -numa node,memdev=ram1

It also hangs when trying to initialize the EFI subsystem in general, e.g 'efidebug memmap', probably because some EFI code is relocated on the region you exlcuded.

Thanks
/Ilias

Quoted 38 lines
---
 lib/efi_loader/efi_memory.c | 4 ++--
 lib/lmb.c                   | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
lib/efi_loader/efi_memory.c +2 -2
index f07cc39b157..57691d15758 100644
@@ -915,8 +915,8 @@ static void add_u_boot_and_runtime(void)
        /* Add U-Boot */
        uboot_start = ((uintptr_t)map_sysmem(gd->start_addr_sp, 0) -
                       uboot_stack_size) & ~EFI_PAGE_MASK;
-       uboot_pages = ((uintptr_t)map_sysmem(gd->ram_top - 1, 0) -
-                      uboot_start + EFI_PAGE_MASK) >> EFI_PAGE_SHIFT;
+       uboot_pages = ((gd->relocaddr - uboot_start) + EFI_PAGE_MASK) >>
+                     EFI_PAGE_SHIFT;
        efi_update_memory_map(uboot_start, uboot_pages, EFI_BOOT_SERVICES_CODE,
                              false, false);
 #if defined(__aarch64__)
lib/lmb.c +1 -2
index 542bb11dcf5..0df8157db7f 100644
@@ -540,7 +540,7 @@ static void lmb_reserve_uboot_region(void)
        ulong pram = 0;
 
        rsv_start = gd->start_addr_sp - CONFIG_STACK_SIZE;
-       end = gd->ram_top;
+       end = gd->relocaddr;
 
        /*
         * Reserve memory from aligned address below the bottom of U-Boot stack
--

2.53.0

#19 lore raw

Re: [PATCH 6/6] memory: reserve from start_addr_sp to relocaddr

Quoted 23 lines · Ilias Apalodimas
Quoted 7 lines · On Thu, 2 Apr 2026 at 03:14, <[email protected]> wrote:

From: Randolph Sapp <[email protected]>

Reserve the memory from gd->start_addr_sp - CONFIG_STACK_SIZE to gd->relocaddr instead of gd->ram_top. This allows platform specific relocation addresses to work without unnecessarily painting over a large range.

This hangs when trying to boot with UEFI on QEMU aarch64. With the qemu_arm64_lwip_defconfig and this command line

qemu-system-aarch64 -m 8192 -smp 2 -nographic -cpu cortex-a57 \
    -machine virt,secure=off \
    -bios u-boot.bin \
    -device virtio-rng-pci \
    -drive id=os,if=none,file=my.iso \
    -device virtio-blk-device,drive=os \
    -object memory-backend-ram,id=ram0,size=4G \
    -object memory-backend-ram,id=ram1,size=4G \
    -numa node,memdev=ram0 \
    -numa node,memdev=ram1
Quoted 6 lines

It also hangs when trying to initialize the EFI subsystem in general, e.g 'efidebug memmap', probably because some EFI code is relocated on the region you exlcuded.

Thanks
/Ilias

Thanks for the QEMU testcase. I see now there are a few cases where gd->relocaddr can actually be decremented, meaning we won't actually cover the entire u-boot memory region. I've added a new global data struct member (end_addr_sp, but if anyone has any better name feel free to comment) to track the original value of relocaddr set by setup_dest_addr and it works like a charm.

I also realized that means I could cut out the PRAM and all of that bank specific logic in lmb_reserve_uboot_region now, so that's cool.

Quoted 38 lines
Quoted 38 lines
---
 lib/efi_loader/efi_memory.c | 4 ++--
 lib/lmb.c                   | 2 +-
 2 files changed, 3 insertions(+), 3 deletions(-)
lib/efi_loader/efi_memory.c +2 -2
index f07cc39b157..57691d15758 100644
@@ -915,8 +915,8 @@ static void add_u_boot_and_runtime(void)
        /* Add U-Boot */
        uboot_start = ((uintptr_t)map_sysmem(gd->start_addr_sp, 0) -
                       uboot_stack_size) & ~EFI_PAGE_MASK;
-       uboot_pages = ((uintptr_t)map_sysmem(gd->ram_top - 1, 0) -
-                      uboot_start + EFI_PAGE_MASK) >> EFI_PAGE_SHIFT;
+       uboot_pages = ((gd->relocaddr - uboot_start) + EFI_PAGE_MASK) >>
+                     EFI_PAGE_SHIFT;
        efi_update_memory_map(uboot_start, uboot_pages, EFI_BOOT_SERVICES_CODE,
                              false, false);
 #if defined(__aarch64__)
lib/lmb.c +1 -2
index 542bb11dcf5..0df8157db7f 100644
@@ -540,7 +540,7 @@ static void lmb_reserve_uboot_region(void)
        ulong pram = 0;
 
        rsv_start = gd->start_addr_sp - CONFIG_STACK_SIZE;
-       end = gd->ram_top;
+       end = gd->relocaddr;
 
        /*
         * Reserve memory from aligned address below the bottom of U-Boot stack
--

2.53.0